2 * Copyright (c) 2011-2014, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 #include <linux/nvme.h>
18 #include <linux/cdev.h>
19 #include <linux/pci.h>
20 #include <linux/kref.h>
21 #include <linux/blk-mq.h>
22 #include <linux/lightnvm.h>
23 #include <linux/sed-opal.h>
24 #include <linux/fault-inject.h>
25 #include <linux/rcupdate.h>
27 extern unsigned int nvme_io_timeout
;
28 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
30 extern unsigned int admin_timeout
;
31 #define ADMIN_TIMEOUT (admin_timeout * HZ)
33 #define NVME_DEFAULT_KATO 5
34 #define NVME_KATO_GRACE 10
36 extern struct workqueue_struct
*nvme_wq
;
37 extern struct workqueue_struct
*nvme_reset_wq
;
38 extern struct workqueue_struct
*nvme_delete_wq
;
46 * List of workarounds for devices that required behavior not specified in
51 * Prefers I/O aligned to a stripe size specified in a vendor
52 * specific Identify field.
54 NVME_QUIRK_STRIPE_SIZE
= (1 << 0),
57 * The controller doesn't handle Identify value others than 0 or 1
60 NVME_QUIRK_IDENTIFY_CNS
= (1 << 1),
63 * The controller deterministically returns O's on reads to
64 * logical blocks that deallocate was called on.
66 NVME_QUIRK_DEALLOCATE_ZEROES
= (1 << 2),
69 * The controller needs a delay before starts checking the device
70 * readiness, which is done by reading the NVME_CSTS_RDY bit.
72 NVME_QUIRK_DELAY_BEFORE_CHK_RDY
= (1 << 3),
75 * APST should not be used.
77 NVME_QUIRK_NO_APST
= (1 << 4),
80 * The deepest sleep state should not be used.
82 NVME_QUIRK_NO_DEEPEST_PS
= (1 << 5),
85 * Supports the LighNVM command set if indicated in vs[1].
87 NVME_QUIRK_LIGHTNVM
= (1 << 6),
90 * Set MEDIUM priority on SQ creation
92 NVME_QUIRK_MEDIUM_PRIO_SQ
= (1 << 7),
95 * Ignore device provided subnqn.
97 NVME_QUIRK_IGNORE_DEV_SUBNQN
= (1 << 8),
101 * Common request structure for NVMe passthrough. All drivers must have
102 * this structure as the first member of their request-private data.
104 struct nvme_request
{
105 struct nvme_command
*cmd
;
106 union nvme_result result
;
110 struct nvme_ctrl
*ctrl
;
114 * Mark a bio as coming in through the mpath node.
116 #define REQ_NVME_MPATH REQ_DRV
119 NVME_REQ_CANCELLED
= (1 << 0),
120 NVME_REQ_USERCMD
= (1 << 1),
123 static inline struct nvme_request
*nvme_req(struct request
*req
)
125 return blk_mq_rq_to_pdu(req
);
128 static inline u16
nvme_req_qid(struct request
*req
)
132 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req
)) + 1;
135 /* The below value is the specific amount of delay needed before checking
136 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
137 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
140 #define NVME_QUIRK_DELAY_AMOUNT 2300
142 enum nvme_ctrl_state
{
145 NVME_CTRL_ADMIN_ONLY
, /* Only admin queue live */
147 NVME_CTRL_CONNECTING
,
154 enum nvme_ctrl_state state
;
157 struct mutex scan_lock
;
158 const struct nvme_ctrl_ops
*ops
;
159 struct request_queue
*admin_q
;
160 struct request_queue
*connect_q
;
164 struct blk_mq_tag_set
*tagset
;
165 struct blk_mq_tag_set
*admin_tagset
;
166 struct list_head namespaces
;
167 struct rw_semaphore namespaces_rwsem
;
168 struct device ctrl_device
;
169 struct device
*device
; /* char device */
171 struct work_struct reset_work
;
172 struct work_struct delete_work
;
174 struct nvme_subsystem
*subsys
;
175 struct list_head subsys_entry
;
177 struct opal_dev
*opal_dev
;
196 atomic_t abort_limit
;
206 unsigned int shutdown_timeout
;
209 unsigned long quirks
;
210 struct nvme_id_power_state psd
[32];
211 struct nvme_effects_log
*effects
;
212 struct work_struct scan_work
;
213 struct work_struct async_event_work
;
214 struct delayed_work ka_work
;
215 struct nvme_command ka_cmd
;
216 struct work_struct fw_act_work
;
217 unsigned long events
;
219 #ifdef CONFIG_NVME_MULTIPATH
220 /* asymmetric namespace access: */
225 struct mutex ana_lock
;
226 struct nvme_ana_rsp_hdr
*ana_log_buf
;
228 struct timer_list anatt_timer
;
229 struct work_struct ana_work
;
232 /* Power saving configuration */
233 u64 ps_max_latency_us
;
249 struct nvmf_ctrl_options
*opts
;
251 struct page
*discard_page
;
252 unsigned long discard_page_busy
;
255 struct nvme_subsystem
{
259 * Because we unregister the device on the last put we need
260 * a separate refcount.
263 struct list_head entry
;
265 struct list_head ctrls
;
266 struct list_head nsheads
;
267 char subnqn
[NVMF_NQN_SIZE
];
270 char firmware_rev
[8];
277 * Container structure for uniqueue namespace identifiers.
286 * Anchor structure for namespaces. There is one for each namespace in a
287 * NVMe subsystem that any of our controllers can see, and the namespace
288 * structure for each controller is chained of it. For private namespaces
289 * there is a 1:1 relation to our namespace structures, that is ->list
290 * only ever has a single entry for private namespaces.
292 struct nvme_ns_head
{
293 struct list_head list
;
294 struct srcu_struct srcu
;
295 struct nvme_subsystem
*subsys
;
297 struct nvme_ns_ids ids
;
298 struct list_head entry
;
301 #ifdef CONFIG_NVME_MULTIPATH
302 struct gendisk
*disk
;
303 struct bio_list requeue_list
;
304 spinlock_t requeue_lock
;
305 struct work_struct requeue_work
;
307 struct nvme_ns __rcu
*current_path
[];
311 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
312 struct nvme_fault_inject
{
313 struct fault_attr attr
;
314 struct dentry
*parent
;
315 bool dont_retry
; /* DNR, do not retry */
316 u16 status
; /* status code */
321 struct list_head list
;
323 struct nvme_ctrl
*ctrl
;
324 struct request_queue
*queue
;
325 struct gendisk
*disk
;
326 #ifdef CONFIG_NVME_MULTIPATH
327 enum nvme_ana_state ana_state
;
330 struct list_head siblings
;
331 struct nvm_dev
*ndev
;
333 struct nvme_ns_head
*head
;
342 #define NVME_NS_REMOVING 0
343 #define NVME_NS_DEAD 1
344 #define NVME_NS_ANA_PENDING 2
347 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
348 struct nvme_fault_inject fault_inject
;
353 struct nvme_ctrl_ops
{
355 struct module
*module
;
357 #define NVME_F_FABRICS (1 << 0)
358 #define NVME_F_METADATA_SUPPORTED (1 << 1)
359 #define NVME_F_PCI_P2PDMA (1 << 2)
360 int (*reg_read32
)(struct nvme_ctrl
*ctrl
, u32 off
, u32
*val
);
361 int (*reg_write32
)(struct nvme_ctrl
*ctrl
, u32 off
, u32 val
);
362 int (*reg_read64
)(struct nvme_ctrl
*ctrl
, u32 off
, u64
*val
);
363 void (*free_ctrl
)(struct nvme_ctrl
*ctrl
);
364 void (*submit_async_event
)(struct nvme_ctrl
*ctrl
);
365 void (*delete_ctrl
)(struct nvme_ctrl
*ctrl
);
366 int (*get_address
)(struct nvme_ctrl
*ctrl
, char *buf
, int size
);
367 void (*stop_ctrl
)(struct nvme_ctrl
*ctrl
);
370 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
371 void nvme_fault_inject_init(struct nvme_ns
*ns
);
372 void nvme_fault_inject_fini(struct nvme_ns
*ns
);
373 void nvme_should_fail(struct request
*req
);
375 static inline void nvme_fault_inject_init(struct nvme_ns
*ns
) {}
376 static inline void nvme_fault_inject_fini(struct nvme_ns
*ns
) {}
377 static inline void nvme_should_fail(struct request
*req
) {}
380 static inline int nvme_reset_subsystem(struct nvme_ctrl
*ctrl
)
382 if (!ctrl
->subsystem
)
384 return ctrl
->ops
->reg_write32(ctrl
, NVME_REG_NSSR
, 0x4E564D65);
387 static inline u64
nvme_block_nr(struct nvme_ns
*ns
, sector_t sector
)
389 return (sector
>> (ns
->lba_shift
- 9));
392 static inline void nvme_end_request(struct request
*req
, __le16 status
,
393 union nvme_result result
)
395 struct nvme_request
*rq
= nvme_req(req
);
397 rq
->status
= le16_to_cpu(status
) >> 1;
399 /* inject error when permitted by fault injection framework */
400 nvme_should_fail(req
);
401 blk_mq_complete_request(req
);
404 static inline void nvme_get_ctrl(struct nvme_ctrl
*ctrl
)
406 get_device(ctrl
->device
);
409 static inline void nvme_put_ctrl(struct nvme_ctrl
*ctrl
)
411 put_device(ctrl
->device
);
414 void nvme_complete_rq(struct request
*req
);
415 bool nvme_cancel_request(struct request
*req
, void *data
, bool reserved
);
416 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
417 enum nvme_ctrl_state new_state
);
418 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
);
419 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
);
420 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
);
421 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
422 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
);
423 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
);
424 void nvme_start_ctrl(struct nvme_ctrl
*ctrl
);
425 void nvme_stop_ctrl(struct nvme_ctrl
*ctrl
);
426 void nvme_put_ctrl(struct nvme_ctrl
*ctrl
);
427 int nvme_init_identify(struct nvme_ctrl
*ctrl
);
429 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
);
431 int nvme_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
, size_t len
,
434 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
, __le16 status
,
435 volatile union nvme_result
*res
);
437 void nvme_stop_queues(struct nvme_ctrl
*ctrl
);
438 void nvme_start_queues(struct nvme_ctrl
*ctrl
);
439 void nvme_kill_queues(struct nvme_ctrl
*ctrl
);
440 void nvme_unfreeze(struct nvme_ctrl
*ctrl
);
441 void nvme_wait_freeze(struct nvme_ctrl
*ctrl
);
442 void nvme_wait_freeze_timeout(struct nvme_ctrl
*ctrl
, long timeout
);
443 void nvme_start_freeze(struct nvme_ctrl
*ctrl
);
445 #define NVME_QID_ANY -1
446 struct request
*nvme_alloc_request(struct request_queue
*q
,
447 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
, int qid
);
448 void nvme_cleanup_cmd(struct request
*req
);
449 blk_status_t
nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
450 struct nvme_command
*cmd
);
451 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
452 void *buf
, unsigned bufflen
);
453 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
454 union nvme_result
*result
, void *buffer
, unsigned bufflen
,
455 unsigned timeout
, int qid
, int at_head
,
456 blk_mq_req_flags_t flags
, bool poll
);
457 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
);
458 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
);
459 int nvme_reset_ctrl(struct nvme_ctrl
*ctrl
);
460 int nvme_reset_ctrl_sync(struct nvme_ctrl
*ctrl
);
461 int nvme_delete_ctrl(struct nvme_ctrl
*ctrl
);
462 int nvme_delete_ctrl_sync(struct nvme_ctrl
*ctrl
);
464 int nvme_get_log(struct nvme_ctrl
*ctrl
, u32 nsid
, u8 log_page
, u8 lsp
,
465 void *log
, size_t size
, u64 offset
);
467 extern const struct attribute_group
*nvme_ns_id_attr_groups
[];
468 extern const struct block_device_operations nvme_ns_head_ops
;
470 #ifdef CONFIG_NVME_MULTIPATH
471 bool nvme_ctrl_use_ana(struct nvme_ctrl
*ctrl
);
472 void nvme_set_disk_name(char *disk_name
, struct nvme_ns
*ns
,
473 struct nvme_ctrl
*ctrl
, int *flags
);
474 void nvme_failover_req(struct request
*req
);
475 void nvme_kick_requeue_lists(struct nvme_ctrl
*ctrl
);
476 int nvme_mpath_alloc_disk(struct nvme_ctrl
*ctrl
,struct nvme_ns_head
*head
);
477 void nvme_mpath_add_disk(struct nvme_ns
*ns
, struct nvme_id_ns
*id
);
478 void nvme_mpath_remove_disk(struct nvme_ns_head
*head
);
479 int nvme_mpath_init(struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
);
480 void nvme_mpath_uninit(struct nvme_ctrl
*ctrl
);
481 void nvme_mpath_stop(struct nvme_ctrl
*ctrl
);
482 void nvme_mpath_clear_current_path(struct nvme_ns
*ns
);
483 struct nvme_ns
*nvme_find_path(struct nvme_ns_head
*head
);
485 static inline void nvme_mpath_check_last_path(struct nvme_ns
*ns
)
487 struct nvme_ns_head
*head
= ns
->head
;
489 if (head
->disk
&& list_empty(&head
->list
))
490 kblockd_schedule_work(&head
->requeue_work
);
493 extern struct device_attribute dev_attr_ana_grpid
;
494 extern struct device_attribute dev_attr_ana_state
;
497 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl
*ctrl
)
502 * Without the multipath code enabled, multiple controller per subsystems are
503 * visible as devices and thus we cannot use the subsystem instance.
505 static inline void nvme_set_disk_name(char *disk_name
, struct nvme_ns
*ns
,
506 struct nvme_ctrl
*ctrl
, int *flags
)
508 sprintf(disk_name
, "nvme%dn%d", ctrl
->instance
, ns
->head
->instance
);
511 static inline void nvme_failover_req(struct request
*req
)
514 static inline void nvme_kick_requeue_lists(struct nvme_ctrl
*ctrl
)
517 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl
*ctrl
,
518 struct nvme_ns_head
*head
)
522 static inline void nvme_mpath_add_disk(struct nvme_ns
*ns
,
523 struct nvme_id_ns
*id
)
526 static inline void nvme_mpath_remove_disk(struct nvme_ns_head
*head
)
529 static inline void nvme_mpath_clear_current_path(struct nvme_ns
*ns
)
532 static inline void nvme_mpath_check_last_path(struct nvme_ns
*ns
)
535 static inline int nvme_mpath_init(struct nvme_ctrl
*ctrl
,
536 struct nvme_id_ctrl
*id
)
538 if (ctrl
->subsys
->cmic
& (1 << 3))
539 dev_warn(ctrl
->device
,
540 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
543 static inline void nvme_mpath_uninit(struct nvme_ctrl
*ctrl
)
546 static inline void nvme_mpath_stop(struct nvme_ctrl
*ctrl
)
549 #endif /* CONFIG_NVME_MULTIPATH */
552 int nvme_nvm_register(struct nvme_ns
*ns
, char *disk_name
, int node
);
553 void nvme_nvm_unregister(struct nvme_ns
*ns
);
554 extern const struct attribute_group nvme_nvm_attr_group
;
555 int nvme_nvm_ioctl(struct nvme_ns
*ns
, unsigned int cmd
, unsigned long arg
);
557 static inline int nvme_nvm_register(struct nvme_ns
*ns
, char *disk_name
,
563 static inline void nvme_nvm_unregister(struct nvme_ns
*ns
) {};
564 static inline int nvme_nvm_ioctl(struct nvme_ns
*ns
, unsigned int cmd
,
569 #endif /* CONFIG_NVM */
571 static inline struct nvme_ns
*nvme_get_ns_from_dev(struct device
*dev
)
573 return dev_to_disk(dev
)->private_data
;
576 int __init
nvme_core_init(void);
577 void __exit
nvme_core_exit(void);